neutron/doc/source/contributor/internals/agent_extensions.rst
Akihiro Motoki 2a47ffd96d Rearrange existing documentation to fit the new standard layout
For more detail, see the doc migration spec.
http://specs.openstack.org/openstack/docs-specs/specs/pike/os-manuals-migration.html

Change-Id: I142a686a3abbe65138a9f3296cd21fc21fbd763a
2017-07-08 05:49:56 +00:00

4.4 KiB

Agent extensions

All reference agents utilize a common extension mechanism that allows for the introduction and enabling of a core resource extension without needing to change agent code. This mechanism allows multiple agent extensions to be run by a single agent simultaneously. The mechanism may be especially interesting to third parties whose extensions lie outside the neutron tree.

Under this framework, an agent may expose its API to each of its extensions thereby allowing an extension to access resources internal to the agent. At layer 2, for instance, upon each port event the agent is then able to trigger a handle_port method in its extensions.

Interactions with the agent API object are in the following order:

  1. The agent initializes the agent API object.
  2. The agent passes the agent API object into the extension manager.
  3. The manager passes the agent API object into each extension.
  4. An extension calls the new agent API object method to receive, for instance, bridge wrappers with cookies allocated.
+-----------+
| Agent API +--------------------------------------------------+
+-----+-----+                                                  |
      |                                   +-----------+        |
      |1                               +--+ Extension +--+     |
      |                                |  +-----------+  |     |
+---+-+-+---+  2  +--------------+  3  |                 |  4  |
|   Agent   +-----+ Ext. manager +-----+--+   ....    +--+-----+
+-----------+     +--------------+     |                 |
                                       |  +-----------+  |
                                       +--+ Extension +--+
                                          +-----------+

Each extension is referenced through a stevedore entry point defined within a specific namespace. For example, L2 extensions are referenced through the neutron.agent.l2.extensions namespace.

The relevant modules are:

  • neutron.agent.agent_extension: This module defines an abstract extension interface for all agent extensions across L2 and L3.
  • neutron.agent.l2.l2_agent_extension:
  • neutron.agent.l3.l3_agent_extension: These modules subclass neutron.agent.agent_extension.AgentExtension and define a layer-specific abstract extension interface.
  • neutron.agent.agent_extensions_manager: This module contains a manager that allows extensions to load themselves at runtime.
  • neutron.agent.l2.l2_agent_extensions_manager:
  • neutron.agent.l3.l3_agent_extensions_manager: Each of these modules passes core resource events to loaded extensions.

Agent API object

Every agent can pass an "agent API object" into its extensions in order to expose its internals to them in a controlled way. To accommodate different agents, each extension may define a consume_api() method that will receive this object.

This agent API object is part of neutron's public interface for third parties. All changes to the interface will be managed in a backwards-compatible way.

At this time, on the L2 side, only the L2 Open vSwitch agent provides an agent API object to extensions. See L2 agent extensions <l2_agent_extensions>. For L3, see L3 agent extensions <l3_agent_extensions>.

The relevant modules are:

  • neutron.agent.agent_extension
  • neutron.agent.agent_extensions_manager
  • neutron.agent.l2.l2_agent_extension_api
  • neutron.agent.l2.l2_agent_extensions_manager
  • neutron.agent.l3.l3_agent_extension_api
  • neutron.agent.l3.l3_agent_extensions_manager